Supervised Isometric Mapping Based Classification Algorithm

In this paper, we propose a novel supervised classification algorithm named Supervised Isometric Mapping Based classification Algorithm (SIMBA). The main idea of SIMBA is to integrate the supervised information into the well-known ISOmetric MAPping (ISOMAP) manifold learning algorithm and classify the transformed data in a low-dimensional feature space. By virtue of the integrated supervised information, the manifold mapping becomes more discriminative, thus the classification performance can be improved. SIMBA can deal with complex high-dimensional data lying on an intrinsically low-dimensional manifold, but only has one free parameter, which is the number of nearest neighbors. Sufficient experiment results demonstrate that SIMBA shows higher classification accuracy on real-world datasets than the state-of-the-art support vector machine classifier.